Citable content: how to get cited in AI search

How to get cited in AI search comes down to the passage, not the page. ChatGPT, Perplexity and Google's AI features quote a few sentences under one heading, so each of those sentences has to name its subject, carry its own number and source, and answer something without help from the text around it. Write definitions in one sentence, keep facts and their sources together, and make list items complete sentences.

This post is the sentence-level editing pass. Access, crawlers and brand mentions are in how to get cited by ChatGPT, and the section-level habits of question headings, answer-first openings and tables are in how to write passages AI can quote. Below are the rewrites those posts don't cover.

How to get cited in AI search with passages that stand alone

A citable passage still answers its heading after you delete everything above it. That is the test I run on every section. Copy the heading and its paragraphs into an empty document and read them cold. If you have to scroll back up to know what "the plan", "this approach" or "the second option" means, an AI answer quoting that passage has the same problem.

Here is a typical section from a made-up payroll product, before and after the test:

Before:
<h2>Pricing</h2>
As we covered above, it depends on team size. The basic tier starts
at a low monthly rate, and the bigger one adds the features most
growing companies need.

After:
<h2>How much does Acme Payroll cost?</h2>
Acme Payroll costs $6 per employee per month on the Starter plan and
$11 per employee per month on the Growth plan, both with a $40 monthly
base fee. The Growth plan adds time tracking and payroll for
contractors in 30 countries.

The before version fails four ways. The heading isn't a question anyone asks, "as we covered above" points outside the passage, "it" has no subject, and "a low monthly rate" gives nothing to quote.

Write definitions in one sentence

A definition passage opens with "X is a [category] that [what makes it different]" and finishes the idea in that sentence. If an AI answer quotes one sentence of your definition, make it the complete one.

Before:
<h2>What is net revenue retention?</h2>
NRR is one of the metrics investors care about most, and for good
reason. When people talk about it, they usually mean how well you
hold on to and grow your existing accounts over time.

After:
<h2>What is net revenue retention?</h2>
Net revenue retention (NRR) is the percentage of recurring revenue a
company keeps from its existing customers over a period, usually 12
months, after counting upgrades, downgrades and cancellations. An NRR
above 100% means existing customers now pay more than they did at the
start of the period, before any new sales.

Spell out the term and its abbreviation together the first time, because people ask both ways. Put the category word early ("percentage", "file", "tag", "plugin"). Then add one sentence that tells the reader how to read the value. Save history and opinions for the paragraphs after.

Give every sentence its own subject

Write each sentence so its subject is a name, not a pointer. The leading "It" on a section is the obvious case, but an AI answer can quote a single sentence from the middle of a passage, and that sentence loses its neighbors too.

Before:
The former runs on the 25th. This means employees see it two days
later. It can be changed in settings, but the latter cannot.

After:
Acme Payroll runs monthly payroll on the 25th, and employees see the
deposit two business days later. Admins can move the monthly run date
in Settings > Pay schedule. Off-cycle bonus payments always run on
the day an admin approves them.

Watch for "the former", "the latter", "here", "the above", "this means" and "both". Each one ties a sentence to another sentence.

Pick one name for each thing and keep it. A section that calls the same feature "pay runs", "payroll cycles" and "the schedule" reads like three features, and a question asking about "pay run dates" matches only a third of it. Writers are taught to vary words. In a passage meant for quoting, repeat the exact term.

Keep the number, the date and the source in one sentence

Put a statistic, the date it applies to and who published it in the same sentence. A footnote marker like [3] or a "Sources" list at the bottom of the page detaches the claim from its evidence the moment one sentence gets quoted.

Before:
FAQ rich results barely show up anymore [3].

After:
Google stopped showing FAQ rich results for every site on May 7, 2026,
according to the Search Central documentation changelog.

On the live page, link the source inside that sentence, as with the Search Central changelog. Write "as of October 2026" instead of "currently". A quoted sentence carries no publish date with it, so "this year" could mean any year to the person reading the answer.

Write lists and tables that make sense when lifted

Make every list item a complete sentence that names its subject. An AI answer can reuse a list, or a single item from it, and a fragment like "Faster setup" or "See step 2" means nothing on its own.

Before:
- Import
- Then map it
- Done in about a day

After:
1. Export your employee list from your current payroll provider as a CSV file.
2. Upload the CSV file in Acme Payroll under People > Import and match each column to an Acme field.
3. Run one test payroll before your first real pay date. Most teams finish setup in one working day.

Start each step with a verb, name the screen or menu, and never write "repeat the above".

For tables, put the unit in the header ("Price per employee, USD") rather than in every cell, and write one prose sentence above the table that states its main conclusion. If an answer quotes only the text around the table, that sentence still answers the question.

Check which of your passages AI can quote

Citability Analyzer splits a page's main content into sections at each h1, h2 and h3 and checks every section. It reports the lead sentence, whether that sentence is a direct answer of 50 words or fewer without filler like "In this section", whether the passage opens with a pronoun such as "It is" or points elsewhere with "as mentioned above", and how many statistics, attributed quotes and outbound sources it holds. It also flags pages with no subheadings, few question headings and sections over 400 words.

You get a 0 to 100 score for English pages with at least 150 words of prose. The tool reads the HTML your server returns, skips text inside tables and code blocks, and checks only the first sentence for a leading pronoun, so read the middle sentences yourself. It doesn't rewrite anything or predict citations. A run costs 5 credits.